CN112637094A - Multi-user MIMO receiving method based on model-driven deep learning - Google Patents

Multi-user MIMO receiving method based on model-driven deep learning Download PDF

Info

Publication number
CN112637094A
CN112637094A CN202011490488.2A CN202011490488A CN112637094A CN 112637094 A CN112637094 A CN 112637094A CN 202011490488 A CN202011490488 A CN 202011490488A CN 112637094 A CN112637094 A CN 112637094A
Authority
CN
China
Prior art keywords
signal
training
network
neural network
deep learning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
CN202011490488.2A
Other languages
Chinese (zh)
Inventor
徐友云
李大鹏
蒋锐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Nanyou Communication Network Industry Research Institute Co ltd
Nanjing Ai Er Win Technology Co ltd
Original Assignee
Nanjing Nanyou Communication Network Industry Research Institute Co ltd
Nanjing Ai Er Win Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Nanyou Communication Network Industry Research Institute Co ltd, Nanjing Ai Er Win Technology Co ltd filed Critical Nanjing Nanyou Communication Network Industry Research Institute Co ltd
Priority to CN202011490488.2A priority Critical patent/CN112637094A/en
Publication of CN112637094A publication Critical patent/CN112637094A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/0413—MIMO systems
    • H04B7/0452—Multi-user MIMO systems
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • H04L25/0254—Channel estimation channel estimation algorithms using neural network algorithms

Abstract

The invention discloses a MIMO receiving method based on model-driven deep learning, which comprises the following steps: s1, pilot signal y to be receivedpInputting the signal into a least square channel estimator to carry out initial channel estimation, and then sending the estimation result obtained by the least square to a fully connected neural network to obtain more accurate channel information
Figure DDA0002840544050000011
S2, data signal y to be receiveddAnd the channel information obtained at S1
Figure DDA0002840544050000012
The initial channel detection result is obtained as the input of the zero forcing signal detector, and then the initial channel detection result is sent into a neural network of an expansion projection gradient algorithm to further improve the signal detection result(ii) a And S3, obtaining the optimal network parameters through training of the neural network in channel estimation and signal detection. The invention greatly reduces the training parameters required by the network, improves the training speed and has high detection performance by combining the deep neural network with the communication model.

Description

Multi-user MIMO receiving method based on model-driven deep learning
Technical Field
The invention relates to a signal receiving method in the technical field of wireless communication, in particular to a combined channel estimation and signal detection method of a multi-input multi-output (MIMO) wireless signal based on model-driven deep learning.
Background
With the advent of the 5G network era, high data rate and high quality communication systems have become the research target of mobile communication. MIMO achieves high rate transmission of information through a kind of multiplexing of channel resources. The technology can improve the system channel capacity by times on the premise of not increasing the frequency spectrum resources and the antenna transmitting power. MIMO has become one of the key technologies for next-generation wireless communication. The performance of the MIMO receiving algorithm directly affects the signal quality of the MIMO system. How to improve the performance of the MIMO receiver has been a hot topic of research.
In recent years, artificial intelligence technology has entered into various fields of life and work. With the improvement of deep learning theory and the rapid improvement of GPU data processing capability in recent years, deep learning has been widely applied to the fields of computer vision, natural language processing, speech recognition, etc. as a key technology of artificial intelligence, and has achieved remarkable results. Meanwhile, in the field of wireless communication, people are also gradually paying attention to a new technology of deep learning. In the prior art, the deep learning technique has been used to solve the problems of beamforming, channel estimation and signal detection. Currently, wireless communication methods based on deep learning are divided into two mainstream methods of data driving and model driving. The data-driven deep learning method treats a certain module of communication as a black box and replaces the module with a deep learning network, wherein the deep learning network utilized by the method comprises a fully-connected network (DNN), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN) and the like. For example, Hao Ye et al, published in IEEE Wireless Communications Letters feb.2018, pp.114-117 (promulgated Wireless communication in the institute of electrical and electronics engineers, 2.2018, page 114, 117), entitled "Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems (Deep Learning capability for Channel Estimation and Signal Detection in orthogonal frequency division multiplexing Systems)", which adopts a data-driven method to train the Deep network model through a large amount of training data, and finally realizes accurate Channel Estimation and Signal Detection for the OFDM system. Model-driven is to build networks based on knowledge and mechanisms known in the field of communications. Through search, Neev Samuel et al published a article entitled "Deep MIMO detection" in 2017IEEE 18th International work on Signal Processing Advances in Wireless Communications (SPAWC)2017, pp.1-5 (IEEE 18th International Workshop on Wireless communication Signal Processing (SPAWC)2017, pp.1-5). The method adopts a mode of driving deep learning by a model, adopts an expanded projection gradient iterative algorithm, adds trainable parameters, obtains a scheme which can be used for MIMO signal detection after training, and obtains better detection performance and faster convergence rate.
It was found through search that "a data model dual-drive MIMO receiver" was invented by seiko et al, south-east university (publication No.: CN 109391315A). The invention discloses a data model dual-drive MIMO receiver, which is formed by connecting networks with the same structure in series of T layers, wherein each layer of network comprises a minimum mean square error de-noising device and a linear estimator; taking the channel state information and the received signal as the input of each layer network, wherein the error variance vector is obtained by combining the output of the t-layer network with the output of the (t-1) -layer network; the t-level network calculates external information according to the input parameters to be trained, the error variance estimation vector and the linear estimator, calculates a posterior probability mean value by adopting a minimum mean square error de-noising device according to the external information, and simultaneously outputs and transmits the posterior probability mean value to the next level network; and outputting the estimated value of the transmission symbol by the T-layer network. The invention can greatly improve the network performance, realize dynamic update and network self-adaptation, can improve the performance of the receiver and obtain obvious performance gain on the basis of the traditional iterative receiver. In addition, the inventor of the Yongming et al of southeast university also found a power control method of an uplink massive MIMO system based on a low-complexity receiver (publication number: CN 105744613A). The invention discloses a power control method of an uplink large-scale MIMO system based on a low-complexity receiver, wherein in the uplink large-scale MIMO system, a Truncation Polynomial (TPE) receiver is adopted and is used as an inversion matrix in a Minimum Mean Square Error (MMSE) receiver; and the base station acquires statistical channel state information, and jointly optimizes polynomial coefficients of the TPE receiver and the uplink transmission power of the user by adopting an iterative algorithm. The invention effectively avoids the inversion operation of the large-dimensional matrix of the MMSE receiver, reduces the complexity and can approach the MMSE receiver in performance.
However, the current MIMO receiving method based on deep learning is based on the condition that the channel information is completely known, and does not take the error generated by channel estimation into account, and the training parameters are too many, and the training time is long, so that the method cannot adapt to the condition of variable environment.
The invention content is as follows:
the present invention aims to overcome the defects of the prior art, and provides a MIMO receiving method for model-driven deep learning, which does not rely on a large amount of data drive completely, and further improves the receiving performance and reduces the training time by combining with the MIMO signal processing receiving knowledge, and completes channel estimation and signal detection at the same time.
The invention is realized by the following technical scheme:
a MIMO receiving method of model-driven deep learning comprises the following steps:
s1, pilot signal y to be receivedpInputting the channel estimation result to Least Square (LS) channel estimator for initial channel estimation, and sending the estimation result obtained by the LS channel estimator to a fully connected neural network to obtain more accurate channel information
Figure BDA0002840544030000031
S2, data signal y to be receiveddAnd the channel information obtained at S1
Figure BDA0002840544030000032
The initial channel detection result is obtained as the input of a Zero Forcing (ZF) signal detector, and then the initial channel detection result is sent into a neural network of an expansion projection gradient algorithm to further improve the signal detection result.
And S3, obtaining the optimal network parameters through training of the neural network in channel estimation and signal detection.
Further, S1 specifically includes:
s11, the received signal is composed of a pilot signal and a data signal, i.e. y ═ yp,yd) Wherein y ispIs a received pilot signal, ydIs a received data signal. Will receive a pilot signal ypAnd local pilot signal xpInputting the result to a least square estimator (LS)
Figure BDA0002840544030000033
Wherein the initialization result
Figure BDA0002840544030000034
The formula adopted is as follows:
Figure BDA0002840544030000035
s12, initializing the least square estimator
Figure BDA0002840544030000036
Deep neural network (net) into first full connection1) Wherein the deep neural network has the expression:
O＝net1(I)＝f0(W0fL(WLfL-1(...f1(W1I+b1)...)+bL)+b0)
where I and O are input and output data of the deep neural network, respectively, Wi、biAnd fi(i ═ 0, 1., L) are the weights, biases, and activation functions, respectively, for the i-th layer neural network, where L is the number of layers in the hidden layer. Thus passing through the net1Obtaining channel information
Figure BDA0002840544030000037
Can be expressed as:
Figure BDA0002840544030000038
further, S2 specifically includes:
s21, data signal y to be receiveddAnd estimated channel information
Figure BDA0002840544030000041
Signal detection result as initialized in ZF signal detector
Figure BDA0002840544030000042
Wherein the zero-forcing equalization result
Figure BDA0002840544030000043
Can be expressed as:
Figure BDA0002840544030000044
s22, estimating the channel
Figure BDA0002840544030000045
Received new track information ydAnd compel equalization results
Figure BDA0002840544030000046
Inputting a network through deep learning2And get more accurate outputAnd (6) discharging. Net2The network is formed by connecting K-layer networks in series by expanding a projection gradient algorithm, and the internal structures of each layer are the same except that learnable parameters of each layer are different. The input of the i-th layer network layer is the output x of the i-1 th layerd,i-1Receiving signal ydAnd channel information
Figure BDA0002840544030000047
Wherein, in the first layer, the input is the initialization result
Figure BDA0002840544030000048
Receiving signal ydAnd channel information
Figure BDA0002840544030000049
net2The implementation process of the ith layer is as follows:
Figure BDA00028405440300000410
Figure BDA00028405440300000411
Figure BDA00028405440300000412
in the formula, (.)TRepresenting a transposition, tanh being a hyperbolic function, ζi、θiAnd gammaiTo learn parameters.
Further, S3 specifically includes:
s31, channel estimation module deep learning network net1The loss function is set to a squared error loss (MSE):
Figure BDA00028405440300000413
wherein χ is the training set of the data model generated offline, | χ | represents the size of the training set.
S32, signal detection module deep learning network net2The training of (1) adopts a step-by-step training mode, namely, the step-by-step training refers to training for multiple times, the number of layers of each training is gradually increased, the training result of the previous training is used as the initialization result of the next training, wherein the training loss function of the ith round is set as square error loss (MSE):
Figure BDA00028405440300000414
wherein ν is the module data training set generated offline, and | ν | represents the size of the training set.
Drawings
FIG. 1 is a schematic flow chart of a MIMO system according to an embodiment of the present invention;
FIG. 2 is a block diagram of a deep learning based channel estimation module according to the present invention;
FIG. 3 is a block diagram of a deep learning based signal detection module according to the present invention.
Detailed Description
The following is a detailed description of the embodiments of the present invention, which is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the protection authority of the present invention is not limited to the following embodiments.
The embodiment is realized by the following steps:
fig. 1 is a flow chart of a MIMO system according to an embodiment, in which pilot signals and data channels are transmitted in one frame, assuming that channel information remains unchanged during a coherent time. The constellation modulation mode of the transmitted data is QPSK, the transmitting end is composed of a plurality of users with single antenna, and the receiving end is a base station with a plurality of antennas. The transmission signal is transmitted through a channel and then received at a receiving end. The method specifically comprises the following steps:
(1) channel estimation
The received signal is composed of pilot signal and dataSignal composition, i.e. y ═ yp,yd) Wherein y ispIs a received pilot signal, ydIs a received data signal. Will receive a pilot signal ypAnd local pilot signal xpInputting the result to a Least Squares (LS) channel estimator for initialization
Figure BDA0002840544030000051
Wherein the initialization result
Figure BDA0002840544030000052
The formula adopted is as follows:
Figure BDA0002840544030000053
initializing the result of least square estimator
Figure BDA0002840544030000054
Deep neural network (net) into first full connection1) Wherein the deep neural network has the expression:
O＝net1(I)＝f0(W0fL(WLfL-1(...f1(W1I+b1)...)+bL)+b0)
where I and O are input and output data of the deep neural network, respectively, Wi、biAnd fi(i ═ 0, 1., L) are the weights, biases, and activation functions, respectively, for the i-th layer neural network, where L is the number of layers in the hidden layer. Thus through the net1More accurate channel information is obtained
Figure BDA0002840544030000055
Can be expressed as:
Figure BDA0002840544030000061
(2) signal detection
Data signal y to be receiveddAnd estimated channel information
Figure BDA0002840544030000062
Signal detection results as initialization in Zero Forcing (ZF) signal detector
Figure BDA0002840544030000063
Wherein the zero-forcing equalization result
Figure BDA0002840544030000064
Can be expressed as:
Figure BDA0002840544030000065
channel estimation result
Figure BDA0002840544030000066
Received new track information ydAnd compel equalization results
Figure BDA0002840544030000067
Inputting a network through deep learning2And a more accurate output is obtained. Net2The network is formed by connecting K-layer networks in series by expanding a projection gradient algorithm, and the internal structures of all the layers are the same except that learnable parameters of each layer are different. The input of the i-th layer network layer is the output x of the i-1 th layerd,i-1Receiving signal ydAnd channel information
Figure BDA0002840544030000068
Wherein, in the first layer, the input is the initialization result
Figure BDA0002840544030000069
Receiving signal ydAnd channel information
Figure BDA00028405440300000610
net2The implementation process of the ith layer is as follows:
Figure BDA00028405440300000611
Figure BDA00028405440300000612
Figure BDA00028405440300000613
in the formula, (.)TRepresenting a transposition, tanh being a hyperbolic function, ζi、θiAnd gammaiAre learnable parameters.
(3) Training of deep learning networks
Channel estimation module deep learning network net1The loss function is set to the squared error loss:
Figure BDA00028405440300000614
where χ is the training set generated offline, | χ | represents the size of the training set.
Signal detection module deep learning network net2The training of (1) adopts a step-by-step training mode, so-called layer-by-layer training refers to training for multiple times, the number of layers of each training is gradually increased, the training result of the previous training is used as the initialization result of the next training, wherein the training loss function of the ith round is set as the square error loss:
Figure BDA0002840544030000071
wherein ν is a training set generated offline, and | ν | represents the size of the training set.
During the training process, a random gradient descent (SGD) method, namely an Adam optimization algorithm, is adopted. Batch training is adopted during training, wherein when the channel estimation module is trained, the size of each batch is 500 data sets, 200 rounds of training are performed in total, each round comprises 10 batches, the initial learning rate is set to be 0.001, and the learning rate is reduced by 5 times in every 40 rounds of the following training. In the training signal detection module, the learning rate was set to 0.001, and 1000 rounds were trained for each layer, each round containing 1250 sets of training data.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (4)

1. A multi-user MIMO receiving method based on model-driven deep learning is characterized by comprising the following steps:
s1, pilot signal y to be receivedpInputting the signal into a least square channel estimator to carry out initial channel estimation, and then sending the estimation result obtained by the least square to a fully connected neural network to obtain more accurate channel information
Figure FDA0002840544020000011
S2, data signal y to be receiveddAnd the channel information obtained at S1
Figure FDA0002840544020000012
The initial channel detection result is obtained as the input of the zero forcing signal detector, and then the initial channel detection result is sent into a neural network for expanding a projection gradient algorithm to further improve the signal detection result;
and S3, training the channel estimation and signal detection internal neural network to obtain the optimal network parameters.
2. The multi-user MIMO receiving method based on model-driven deep learning of claim 1, wherein: s1 specifically includes:
s11 shows that the received signal y is (y)p,yd) Wherein y ispIs a received pilot signal, ydIs a received data signal; will receive a pilot signal ypAnd local pilot signal xpInputting the result to a least square estimator for initialization
Figure FDA0002840544020000013
Wherein the initialization result
Figure FDA0002840544020000014
The formula adopted is as follows:
Figure FDA0002840544020000015
s12, initializing the least square estimator
Figure FDA0002840544020000016
Deep neural network (net) into first full connection1) Wherein the deep neural network has the expression:
O＝net1(I)＝f0(W0fL(WLfL-1(...f1(W1I+b1)...)+bL)+b0)
where I and O are input and output data of the deep neural network, respectively, Wi、biAnd fi(i ═ 0, 1.., L) are the weights, biases, and activation functions of the i-th layer neural network, respectively, where L is the number of layers of the hidden layer; thus net1Further obtaining more accurate channel information
Figure FDA0002840544020000017
Can be expressed as:
Figure FDA0002840544020000018
3. the multi-user MIMO receiving method based on model-driven deep learning of claim 1, wherein: s2 specifically includes:
s21, data signal y to be receiveddAnd estimated channel information
Figure FDA0002840544020000021
Signal detection results as initialized in zero forcing signal detector
Figure FDA0002840544020000022
Wherein the zero-forcing equalization result
Figure FDA0002840544020000023
Can be expressed as:
Figure FDA0002840544020000024
s22, estimating the channel
Figure FDA0002840544020000025
Received new track information ydAnd compel equalization results
Figure FDA0002840544020000026
Inputting a network through deep learning2Neutralizing and obtaining more accurate output; net2The network is formed by connecting K-layer networks in series by expanding a projection gradient algorithm, and each layer can learnThe parameters are different, and the internal structures are the same firstly; the input of the i-th layer network layer is the output x of the i-1 th layerd,i -1Receiving signal ydAnd channel information
Figure FDA0002840544020000027
Wherein, in the first layer, the input is the initialization result
Figure FDA0002840544020000028
Receiving signal ydAnd channel information
Figure FDA0002840544020000029
net2The implementation process of the ith layer is as follows:
Figure FDA00028405440200000210
Figure FDA00028405440200000211
Figure FDA00028405440200000212
in the formula, (.)TRepresenting a transposition, tanh being a hyperbolic function, ζi、θiAnd gammaiAre learnable parameters.
4. The multi-user MIMO receiving method based on model-driven deep learning of claim 1, wherein: s3 specifically includes:
s31, channel estimation module deep learning network net1The loss function is set to a squared error loss (MSE):
Figure FDA00028405440200000213
wherein χ is a training set generated under the line, | χ | represents the size of the training set;
s32, signal detection module deep learning network net2The training of (1) adopts a step-by-step training mode, wherein the training loss function of the ith round is set as square error loss (MSE):
Figure FDA0002840544020000031
wherein ν is a training set generated offline, and | ν | represents the size of the training set.
CN202011490488.2A 2020-12-17 2020-12-17 Multi-user MIMO receiving method based on model-driven deep learning Withdrawn CN112637094A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011490488.2A CN112637094A (en) 2020-12-17 2020-12-17 Multi-user MIMO receiving method based on model-driven deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011490488.2A CN112637094A (en) 2020-12-17 2020-12-17 Multi-user MIMO receiving method based on model-driven deep learning

Publications (1)

Publication Number Publication Date
CN112637094A true CN112637094A (en) 2021-04-09

Family

ID=75313949

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011490488.2A Withdrawn CN112637094A (en) 2020-12-17 2020-12-17 Multi-user MIMO receiving method based on model-driven deep learning

Country Status (1)

Country Link
CN (1) CN112637094A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113238227A (en) * 2021-05-10 2021-08-10 电子科技大学 Improved least square phase unwrapping method and system combined with deep learning
CN113676431A (en) * 2021-07-08 2021-11-19 东南大学 Model-driven MIMO-OFDM receiving method without cyclic prefix
CN114337745A (en) * 2022-01-10 2022-04-12 重庆邮电大学 Large-scale MIMO signal detection method for model-driven deep learning
CN114696933A (en) * 2022-04-01 2022-07-01 西安交通大学 AI receiver based on deep learning technology and use method
CN114759997A (en) * 2022-04-08 2022-07-15 山东大学 Dual-drive MIMO system signal detection method based on data model
CN114884775A (en) * 2022-03-31 2022-08-09 南京邮电大学 Deep learning-based large-scale MIMO system channel estimation method
CN114915361A (en) * 2022-05-13 2022-08-16 西安交通大学 Internet of things uplink signal detection method based on small sample learning
CN115395991A (en) * 2022-07-13 2022-11-25 北京信息科技大学 Nonlinear multiple-input multiple-output channel estimation method and estimation system
JP7446387B2 (en) 2022-07-13 2024-03-08 緯創資通股▲ふん▼有限公司 Communication method and communication device

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113238227A (en) * 2021-05-10 2021-08-10 电子科技大学 Improved least square phase unwrapping method and system combined with deep learning
CN113676431A (en) * 2021-07-08 2021-11-19 东南大学 Model-driven MIMO-OFDM receiving method without cyclic prefix
CN114337745A (en) * 2022-01-10 2022-04-12 重庆邮电大学 Large-scale MIMO signal detection method for model-driven deep learning
CN114884775A (en) * 2022-03-31 2022-08-09 南京邮电大学 Deep learning-based large-scale MIMO system channel estimation method
CN114696933A (en) * 2022-04-01 2022-07-01 西安交通大学 AI receiver based on deep learning technology and use method
CN114696933B (en) * 2022-04-01 2023-02-07 西安交通大学 AI receiver based on deep learning technology and use method
CN114759997A (en) * 2022-04-08 2022-07-15 山东大学 Dual-drive MIMO system signal detection method based on data model
CN114915361A (en) * 2022-05-13 2022-08-16 西安交通大学 Internet of things uplink signal detection method based on small sample learning
CN115395991A (en) * 2022-07-13 2022-11-25 北京信息科技大学 Nonlinear multiple-input multiple-output channel estimation method and estimation system
CN115395991B (en) * 2022-07-13 2023-08-25 北京信息科技大学 Nonlinear multi-input multi-output channel estimation method and estimation system
JP7446387B2 (en) 2022-07-13 2024-03-08 緯創資通股▲ふん▼有限公司 Communication method and communication device

Similar Documents

Publication Publication Date Title
CN112637094A (en) Multi-user MIMO receiving method based on model-driven deep learning
Aoudia et al. Model-free training of end-to-end communication systems
CN108650201B (en) Neural network-based channel equalization method, decoding method and corresponding equipment
CN110719239B (en) Data model dual-drive combined MIMO channel estimation and signal detection method
CN112600772B (en) OFDM channel estimation and signal detection method based on data-driven neural network
CN106850013B (en) Signal detection method of uplink large-scale MIMO system
Wang et al. Pilot-assisted channel estimation and signal detection in uplink multi-user MIMO systems with deep learning
CN110336594B (en) Deep learning signal detection method based on conjugate gradient descent method
KR102334011B1 (en) Method and apparatus for limited feedback based on machine learning in wireless communication system
CN108964725B (en) Sparse estimation method of channel parameters in time-varying large-scale MIMO network
CN112217545B (en) Large-scale MIMO system detection model construction method
CN114499601B (en) Large-scale MIMO signal detection method based on deep learning
CN109391315B (en) Data model dual-drive MIMO receiver
CN113541747B (en) Large-scale MIMO detection method, device and storage medium
Imanishi et al. Deep learning-aided iterative detector for massive overloaded MIMO channels
Wang et al. Deep Transfer Learning for Model-Driven Signal Detection in MIMO-NOMA Systems
Ali et al. GFDM transceiver based on ann channel estimation
CN110868244B (en) Low-complexity communication signal detection method based on channel puncture
Huang et al. An efficient ADMM-aided deep learning-based signal detector for uplink massive MIMO
CN111193534B (en) Low-complexity signal detection method in large-scale MIMO system
Al-Baidhani et al. Deep ensemble learning: A communications receiver over wireless fading channels
CN114362884B (en) Single-bit large-scale MIMO receiving method using soft information feedback
CN115801500A (en) High-order large-scale MIMO signal detection method
CN113438190B (en) Neural network training method and device, MIMO equalizer and method and readable medium
Maki et al. A From-Scratch Approach to Deep Unfolding-Aided Bilinear Gaussian Belief Propagation for Correlated Large MIMO

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WW01 Invention patent application withdrawn after publication

Application publication date: 20210409

WW01 Invention patent application withdrawn after publication